Qualcomm Research: Robust positioning from visual-inertial and GPS

Image: GPS World
Image: GPS World

Presented at ION GNSS+, September 2016

GPS positioning in urban scenarios is challenging because of large numbers of non-line-of-sight outlier measurements. We propose a robust positioning algorithm that combines GPS observations with visual-inertial odometry information to handle such outliers. We demonstrate the effectiveness of our algorithm in a simulation scenario with close to 80-percent outliers. In experiments in a mild urban-canyon environment, our approach reduces the 95th percentile horizontal positioning error by 66 percent compared to a GPS-only solution.

Motivation

GPS performance drastically degrades if large parts of the sky are obstructed. This occurs for example in urban-canyon scenarios, where GPS positions may be off by as much as 50 meters. These large positioning errors are prohibitive in applications such as autonomous vehicles and advanced driver assistance systems (ADAS).The large positioning errors in urban canyons are mainly caused by non-line-of-sight (NLOS) observations and multipath effects. Such observations result when the line-of-sight (LOS) path from the receiver to a satellite is blocked, and the receiver instead erroneously tracks a reflected version of the satellite signal.

Summary of Results

We propose a low-cost method to detect and remove such NLOS outliers by combining GPS pseudorange measurements with visual inertial odometry (VIO) measurements. These measurements are complementary: GPS pseudoranges provide absolute positioning information; VIO measurements, constructed from camera frames and inertial measurements, provide high-accuracy relative positioning.

We develop a robust and efficient, tightly-coupled GPS+VIO positioning algorithm, able to work under extremely challenging conditions. For example, in scenarios with close to 80 percent of GPS measurement outliers or with only intermittent satellite visibility. Even under these extreme conditions, the proposed algorithms are able to produce reliable and accurate position estimates.

Problem Setting

The overall positioning system consists of a GPS module and a VIO module. The GPS module provides raw pseudorange and Doppler range-rate measurements. The VIO module consists of a camera along with inertial sensors such as an accelerometer and a gyroscope. The output of the VIO processing engine are vectors of velocities and displacements expressed in the local camera coordinate frame.

We will not go into the details of the VIO design, rather we will use it as a black box that provides us with the velocities. The goal is to integrate the pseudorange measurements across time using the highly accurate velocities from the VIO to detect and discard the measurements corrupted by NLOS errors.

The positioning algorithm consists of two stages. In the first stage, we transform the velocities from the VIO frame of reference to the GPS frame of reference. This requires estimation of the rotation matrix relating the VIO frame and the GPS frame. Once this transformation is completed, the second stage is to perform outlier detection and to estimate the rover position.

This article is tagged with , and posted in From the Magazine, GNSS

1 Comment on "Qualcomm Research: Robust positioning from visual-inertial and GPS"

Trackback | Comments RSS Feed

  1. William K. says:

    Very interesting. The concern about the navigation of autonomous vehicles is particularly important because of how much it is ignored by those touting the “wonderfulness” of self driving vehicles the loudest. My point is that while this added system offers a real improvement, it is neither a trivial addition nor a cheap fix. Clearly it offers real value, but it is still dependant on fairly clear visibility of the surroundings, which is not always the case.